: A packaged version including clean COCO annotations is available on Hugging Face Datasets .
While multiple versions exist under similar names, the original person re-identification benchmark is typically hosted or referenced via academic repositories:
: VIPeR.v1.0.zip (Hosted by UC Santa Cruz). viper dataset download
The VIPeR dataset is relatively small by modern standards but focuses on extreme . Total Identities 632 distinct individuals Total Images 1,264 (2 images per person) Camera Views 2 non-overlapping cameras (Camera A and Camera B) Resolution Low-resolution crops (48x128 pixels) Annotation Manual bounding boxes and viewpoint angle annotations Format BMP images Common Variants (The "Viper" Confusion)
The is a cornerstone benchmark in computer vision, specifically designed for person re-identification (Re-ID) . Introduced in 2008, it remains one of the most widely used—and challenging—datasets for testing how well algorithms can identify the same person across different camera views with varying angles, lighting, and poses. Direct Download Links : A packaged version including clean COCO annotations
: Many research repositories like torchreid and Open-ReID include automated scripts that download and format the dataset directly into your environment. Dataset Specifications
VIPeR † ( viper ) The download link is http://users.soe.ucsc.edu/~manduchi/VIPeR.v1.0.zip. Kimhi/viper · Datasets at Hugging Face Total Identities 632 distinct individuals Total Images 1,264
Because "VIPER" is a popular acronym, several unrelated datasets share the name. Ensure you are downloading the correct one for your research: Datasets — torchreid 1.4.0 documentation - Kaiyang Zhou